potiuk opened a new issue, #68417:
URL: https://github.com/apache/airflow/issues/68417
#59691 ("Fix upstream map index resolution after placeholder expansion") was
reverted in #PR_NUM because it regressed per-index trigger-rule evaluation
inside mapped task groups.
**What #59691 did:** added `_should_use_post_expansion_placeholder` in
`taskinstance.py`, which rewrites a downstream placeholder's upstream
resolution from `map_index=-1` to `0` once the upstream has expanded.
**Why it broke main:** the same helper feeds *both* XCom/arg resolution (the
case #59691 targeted) and trigger-rule dependency evaluation. For a task in a
mapped task group whose upstream is in the *same* group (e.g. `divide(i) >>
report_success(i)`), rewriting the summary instance (`-1`) to upstream index
`0` makes it inherit that single instance's state — so when `divide(0)` fails,
the downstream summary is marked `UPSTREAM_FAILED` and never expands. This
regressed the #50210 fix;
`test_one_failed_trigger_rule_in_mapped_task_group_is_per_index` (and the whole
`Core...Serialization` job) went red on `main`.
**Acceptance criteria for a redo:**
- `test_downstream_placeholder_handles_upstream_post_expansion` (the #59691
scenario) passes, **and**
- `test_one_failed_trigger_rule_in_mapped_task_group_is_per_index` +
`test_one_failed_trigger_rule_runs_on_indirect_failure_in_mapped_task_group`
(the #50210 scenarios) pass.
- The placeholder→0 rewrite must not affect trigger-rule evaluation of
not-yet-expanded summary instances (consider distinguishing the XCom-resolution
caller, or handling the summary instance in `trigger_rule_dep.py`).
cc @SameerMesiah97
---
Drafted-by: Claude Code (Opus 4.8); reviewed by @potiuk before posting
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
